Fastai v2 chat

[fmobrj75 (Fábio Barros)]

Hello!

I am having a hard time trying to make weighted_dataloaders to work. @sgugger, @muellerzr any ideas? Here is the code:

tfms = [[attrgetter("filename"), PILImage.create],
        [attrgetter("target"),Categorize()]]

splits=ColSplitter('is_valid')(df)

dsets = Datasets(df, tfms, splits=splits)

dls = dsets.weighted_dataloaders(wgts=df['wgt'].tolist(),bs=8, source=df,
                        after_item = [Resize(528, method='squish'), ToTensor()],
                        after_batch= [IntToFloatTensor(), 
                                      *aug_transforms(size=528, 
                                                      do_flip=True,
                                                      max_rotate=15,
                                                      max_zoom=1.1,
                                                      max_lighting=0.3,
                                                      max_warp=0.0,
                                                      p_affine=1.0,
                                                      p_lighting=1.0
                                                      ), 
                                      Normalize.from_stats(*imagenet_stats)
                       )

The code work with a normal dataloader, without weights, but with weighted_dataloaders I always get this error:

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-31-90634fcc3c9e> in <module>
----> 1 dls.show_batch()

/srv/conda/envs/saturn/lib/python3.7/site-packages/fastai2/data/core.py in show_batch(self, b, max_n, ctxs, show, unique, **kwargs)
     91             old_get_idxs = self.get_idxs
     92             self.get_idxs = lambda: Inf.zeros
---> 93         if b is None: b = self.one_batch()
     94         if not show: return self._pre_show_batch(b, max_n=max_n)
     95         show_batch(*self._pre_show_batch(b, max_n=max_n), ctxs=ctxs, max_n=max_n, **kwargs)

/srv/conda/envs/saturn/lib/python3.7/site-packages/fastai2/data/load.py in one_batch(self)
    129     def one_batch(self):
    130         if self.n is not None and len(self)==0: raise ValueError(f'This DataLoader does not contain any batches')
--> 131         with self.fake_l.no_multiproc(): res = first(self)
    132         if hasattr(self, 'it'): delattr(self, 'it')
    133         return res

/srv/conda/envs/saturn/lib/python3.7/site-packages/fastcore/utils.py in first(x)
    174 def first(x):
    175     "First element of `x`, or None if missing"
--> 176     try: return next(iter(x))
    177     except StopIteration: return None
    178 

/srv/conda/envs/saturn/lib/python3.7/site-packages/fastai2/data/load.py in __iter__(self)
     95         self.randomize()
     96         self.before_iter()
---> 97         for b in _loaders[self.fake_l.num_workers==0](self.fake_l):
     98             if self.device is not None: b = to_device(b, self.device)
     99             yield self.after_batch(b)

/srv/conda/envs/saturn/lib/python3.7/site-packages/torch/utils/data/dataloader.py in __init__(self, loader)
    379 
    380         self._dataset_fetcher = _DatasetKind.create_fetcher(
--> 381             self._dataset_kind, self._dataset, self._auto_collation, self._collate_fn, self._drop_last)
    382 
    383     def _next_data(self):

/srv/conda/envs/saturn/lib/python3.7/site-packages/torch/utils/data/dataloader.py in create_fetcher(kind, dataset, auto_collation, collate_fn, drop_last)
     39             return _utils.fetch._MapDatasetFetcher(dataset, auto_collation, collate_fn, drop_last)
     40         else:
---> 41             return _utils.fetch._IterableDatasetFetcher(dataset, auto_collation, collate_fn, drop_last)
     42 
     43 

/srv/conda/envs/saturn/lib/python3.7/site-packages/torch/utils/data/_utils/fetch.py in __init__(self, dataset, auto_collation, collate_fn, drop_last)
     19     def __init__(self, dataset, auto_collation, collate_fn, drop_last):
     20         super(_IterableDatasetFetcher, self).__init__(dataset, auto_collation, collate_fn, drop_last)
---> 21         self.dataset_iter = iter(dataset)
     22 
     23     def fetch(self, possibly_batched_index):

/srv/conda/envs/saturn/lib/python3.7/site-packages/fastai2/data/load.py in __iter__(self)
     25         store_attr(self, 'd,pin_memory,num_workers,timeout')
     26 
---> 27     def __iter__(self): return iter(self.d.create_batches(self.d.sample()))
     28 
     29     @property

/srv/conda/envs/saturn/lib/python3.7/site-packages/fastai2/data/load.py in sample(self)
     89 
     90     def sample(self):
---> 91         idxs = self.get_idxs()
     92         return (b for i,b in enumerate(idxs) if i//(self.bs or 1)%self.nw==self.offs)
     93 

/srv/conda/envs/saturn/lib/python3.7/site-packages/fastai2/callback/data.py in get_idxs(self)
     23         if self.n==0: return []
     24         if not self.shuffle: return super().get_idxs()
---> 25         return list(np.random.choice(self.n, self.n, p=self.wgts))
     26 
     27 # Cell

mtrand.pyx in numpy.random.mtrand.RandomState.choice()

ValueError: 'a' and 'p' must have same size

[WaterKnight (David Lacalle Castillo)]

@sgugger at the beginning of fastai2.vision.learner code on github, line 12 is duplicated.

from . import models

[paidion (Jeremiah YS Chung)]

fastbook, 04+mnist_basic, When I run:

#id gradient_descent
#caption The gradient descent process
#alt Graph showing the steps for Gradient Descent
gv('''
init->predict->loss->gradient->step->stop
step->predict[label=repeat]
''')

get the following error:

NameError: name 'gv' is not defined

Any suggestion? Thanks.

[muellerzr (Zachary Mueller)]

Did you import the utils file?

[sgugger]

Thanks, it’s fixed now.

[WaterKnight (David Lacalle Castillo)]

I managed to make segmentation one work like follows:

Torchvision Segmentation Models with FastAI 2

I managed to get it working as follows: import torchvision model=torchvision.models.segmentation.deeplabv3_resnet50(pretrained=False,num_classes=2) model.train() class GetResult(Callback): def after_pred(self): self.learn.pred = self.pred["out"] learn = Learner(dls=dls, model=model, metrics=[Dice(),JaccardCoeff()],wd=1e-2, cbs=GetResult()).to_fp16()

However, I tried to change backbone to resnet34 and I am getting: NotImplementedError: Dilation > 1 not supported in BasicBlock

[paidion (Jeremiah YS Chung)]

I think so.

#hide
from fastai2.vision.all import *
from utils import *
matplotlib.rc('image', cmap='Greys')

Something else missing?

[vrodriguezf (Víctor Rodríguez)]

Hi!

I’m trying to use the WandbCallback in a TabularLearner. However, I’m having the following errors when calling fit_one_cycle:

Could not set wandb config input dimensions
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
/usr/local/lib/python3.6/dist-packages/fastai2/learner.py in fit(self, n_epoch, lr, wd, cbs, reset_opt)
    187             try:
--> 188                 self._do_begin_fit(n_epoch)
    189                 for epoch in range(n_epoch):

29 frames
AttributeError: tfms

During handling of the above exception, another exception occurred:

AttributeError                            Traceback (most recent call last)
/usr/local/lib/python3.6/dist-packages/fastai2/callback/wandb.py in after_fit(self)
     83     def after_fit(self):
     84         self.run = True
---> 85         if self.log_preds: self.remove_cb(self.learn.fetch_preds)
     86 
     87     def _log_config(self):

AttributeError: 'TabularLearner' object has no attribute 'fetch_preds'

Any ideas? Thank you for your awesome help!

[fmobrj75 (Fábio Barros)]

@sgugger It seems to me that it is conflicting with splits. When I create a dataset without splits, the wgts seems to be working. Otherwise it complains about the sizes of the parameters (weighted_dataloaders):

ValueError: 'a' and 'p' must have same size

[sgugger]

I have not designed that particular API, so there is no need to tag me. I have no more knowledge of what is causing your error than you do

[fmobrj75 (Fábio Barros)]

I did not know that. Sorry.

[fmobrj75 (Fábio Barros)]

@boris

@patch @delegates(Datasets.dataloaders) def weighted_dataloaders(self:FilteredBase, wgts, bs=64, **kwargs): xtra_kwargs = [{}] * (self.n_subsets-1) return self.dataloaders(bs=bs, dl_type=WeightedDL, dl_kwargs=({‘wgts’:wgts}, *xtra_kwargs), **kwargs)

Hi Boris.

Do you know how the weights must be passed to the weighted_dataloaders? When I do not have splits, it works fine with an array of weights. But when I have splits, I get the following error:

ValueError: 'a' and 'p' must have same size

[boris (Boris Dayma)]

They should be passed only to the training dataset. They won’t be used on the validation dataset (which will process all the items it has).

[fmobrj75 (Fábio Barros)]

Oh, ok! Thank you very much!

[WaterKnight (David Lacalle Castillo)]

Thank you for fixing them, I found some imports being duplicated too. I opened a pull request.

My solution didn’t work. Sorry, i didn’t say nothing!

[adpq]

I’m trying to run the fastbook notebooks. I installed fastai2 and fastcore as editable installs, but I get the following error when I try to run the first line of the notebook:

ModuleNotFoundError: No module named ‘fastai2’

How can I get fastai2 to work in these notebooks?

[paidion (Jeremiah YS Chung)]

How about? !pip install fastai2

[zlapp]

Is the Monte Carlo Dropout functionality introduced here PR: Ability to use dropout at prediction time (Monte Carlo Dropout) available in fastai v2?

[much_learner (Valeriy)]

What’s the point of having size argument in both Resize in item_tfms and aug_transforms in batch_tfms?

[eniolasonowo]

Goodday , ill like to ask if you had success with it using fastai, i am stuck at trying to train, after passing it to a learner and trying to train i get a no target error.